CIRJE-F-140 Pricing Convertible Bonds with Default Risk: ADuffie-Singleton Approach Akihiko Takahashi The University of Tokyo Takao Kobayashi The University of Tokyo Naruhisa Nakagawa Goldman Sachs Japan Ltd. November 2001 Discussion Papers are a series of manuscripts in their draft form. They are not intended for circulation or distribution except as indicated by the author. For that reason Discussion Papers may not be reproduced or distributed without the written consent of the author.
This paper is forthcoming in the Journal of Fixed Income.
Pricing Convertible Bonds with Default Risk: A Due-Singleton Approach Akihiko Takahashi, Takao Kobayashi y and Naruhisa Nakagawa z Abstract We propose a new method to value convertible bonds(cbs). In particular, we explicitly take default risk into consideration based on Due- Singleton(1999), and provide a consistent and practical method for relative pricing of securities issued by a rm such as CBs, non-convertible corporate bonds and equities. Moreover, we show numerical examples using Japanese CBs' data, and compare our model with other practical models. Associate Professor, Graduate School of Mathematical Sciences, University oftokyo, 3-8-1, Komaba, Meguro-ku, Tokyo, 153-8914, Japan. phone: 81-3-5465-7077. fax:81-3- 5465-7011. e-mail: akihiko@ms.u-tokyo.ac.jp y Professor, Faculty of Economics, University of Tokyo, 7-3-1, Hongo, Bunkyo-ku, Tokyo, 113, Japan. phone: 81-3-5841-5518. fax:81-3-5841-5521. e-mail: kobayashi@e.utokyo.ac.jp z Equity Division, Goldman Sachs (Japan) Ltd., ARK Mori Bldg. 5F, 1-12-32, Akasaka Minato-ku, Tokyo, 107-6005, Japan. phone: 81-3-3589-9247. fax:81-3-3587-9264. e-mail: naruhisa.nakagawa@gs.com 1
1 Introduction 1.1 Convertible Bonds Convertible bonds(cbs), corporate bonds that can be converted to equities are widely issued and traded in the nancial markets of various countries. It is well known that many types of options are embedded in a CB: The most important and basic option is a call option on the equity and the option is usually American or Bermudan type. Call or/and put conditions may be added, and recently CBs whose conversion prices depend on the current stock prices become popular in the market. It is important to determine the underlying state variables to value and hedge such a complicated option because the future value of a CB is subject to several risk sorces. Looking at the structure of the product, we easily notice that the state variables should be the stock price, the default-free interest rate and credibility of the issuer. More essentially, CBs, equities and corporate bonds are all derivatives of the underlying rm value. Hence, if we choose the rm value as a state variable focusing on the capital structure of the rm, we are able to evaluate consistently all the securities issued by the rm. On the other hand, this approach cannot be easily implemented since the rm value and its stochastic process, unobservable in the market, must be specied. It is usually dicult to evaluate CBs based on the rm value by utilizing the stock price and the prices of corporate bonds that are observable in the market because those prices themselves depend on the rm value nonlinearly. Thus, in this paper, we take the underlying stock 2
price as a state variable. In addition to the stock price, we focus on default event as a state variable. CBs issued by "new economy" rms that have low credibility might increase since they can be used in nancing for active growth, which is the case in the United States. Thus, pricing CB with default risk will become more important. We also remark that the variables explaining the process of a default-free rate as in term structure models can be easily added as state variables. Finally, we notice that although pricing CBs without passing through the rm value, we should value CBs, nonconvertible corporate bonds and equities consistently within the model in terms of relative pricing. 1.2 Modeling Default Risk Valuing nancial products with default risk can be separated into two approaches such asstructural Approach(A1) and Reduced-form Approach(A2). A1 regards default as a endogenous event by focusing on the capital structure of the rm. Merton[1974] initiated this approach where he expressed the rm value as a diusion process and characterized default by the event that the value of the rm went below the face value of a debt. In this formulation, corporate bonds are options on the rm value and options on the corporate bonds are compound options on the rm value. However, to value a corporate bond in practice, the debts that are senior to it must be simultaneously valued, which raises computational diculty. The problem is more serious when valuing options because of its compound option feature. To overcome the problem, Longsta-Schwartz[1995] modied the approach 3
in that default was characterized by a stopping time of a rm value to a certain boundary which was common among all the debts of the rm. They also introduced a default-free interest rate as the second state variable emphasizing the relation between default and the interest rate. In practical implementation, however, there are still problems in the model. First, the rm value is not directly tradable and is unobservable in the market, which makes parameters' estimation quite dicult. Second, some argue that the event of default should be modeled by a jump process because diusion processes are not able to explain the empirical observation that there are large credit spreads even just before maturity. (See Madan-Unal[1993] for instance.) Forcusing on the problems, Due-Singleton[1999] and Jarrow- Turnbell[1994] proposed models in A2. They did not explain the event of default endogenously, but characterized it exogenously by a jump process, and derived the arbitrage-free prices of securities subject to common default risk. Further, the parameters of models could be estimated or calibrated more directly from observable prices of trading securities such as corporate bonds. In pricing CBs with default risk, most of existing literatures took the structual approach(a1); Ingersoll[1977] and Brenann-Schwartz[1977,1980] developed models within the class. Recently, Tsiveriotis-Fernandes[1998] proposed a model with exogenously given credit spreads. However, they just used spreads observed in markets without any theoretical consideration, and there may exist some inconsistency in the model because default 4
risk embedded in equities was not taken into account. We will propose a new model in reduced-form approach(a2) to price CBs with default risk. In particular, we explicitly take default risk into consideration based on Due- Singleton[1999], and provide a consistent method for relative pricing of CBs, non-convetible corporate bonds and equities issued by a rm. Moreover, we give numerical examples using Japanese CBs' data, and conrm that our model is valid through comparison with other practical models. From practical view point, an advantage of Due-Singleton[1999] is that a jump term does not explicitly appear in the valuation formula and hence that computational methods developed for diusion models can be utilized. The reason is that they derive the valuation formula for the pre-default values of defaultable assets; the pre-default value is dened by the value whose process is equivalent to the price process of an asset before default, and it can be modeled by a diusion process under some regurality conditions. As a result, the valuation formula is similar to the one for default-free assets except in that the discout rate is not a default-free short-rate, but the default-adjusted short-rate, which is determined by a default-free rate, a default hazard rate and the fractional loss of market value of the claim at default. In this approach, all the securities issued by a company have the same hazard rate while they could have dierent loss rates. We model the hazard rate as a decreasing function of the stock price since the stock price is easily observed and traded most frequently among the securities issued by a company, and it seems natural that the probability of default is negatively related with 5
the level of the stock price. We also assume that the pre-default process of the stock price follows a diusion process noting that the stock price itself is subject to default risk. Then we can obtain a consistent pricing method for any securities by specifying their payos and their fractional loss rates at default under an assumption for the stochastic process of the default-free rate. The organization of the paper is as follows. Section 2 describes our model in the framework of Due and Singleton[1999]. Section 3 shows comparison of our model with other models by using data of the Japanese CB market, and the nal section gives conclusion. 2 A CB model in Due-Singleton Approach Under an appropriate mathematical setting, Due and Singleton[1999] shows that the pre-default value V t of a defaultable security characterized by the nal payo X at time T and the cumulative dividend process fd t :0 t T g, is expressed as V t = E Q t " R e, T t R(u)du X + Z T t e, R s t R(u)du dd s # : (1) Here, R(t) is the default-adjusted discount rate dened by R(t) :=r(t)+l(t)(t) (2) where r(t) is the default-free short rate, L(t) is the fractional loss rate of market value at default, (t) is the default hazard rate, and E t [] denotes the conditional expectation under a risk-neutral measure Q, given available 6
information at time t. Hereafter V t denotes the value of a CB provided that default has not occured by time t; if t is the date of a coupon payment, V t denotes the ex-coupon value. For modeling CBs in this framework, we take the pre-defalut value of the underlying stock, S t as a state variable, and suppose that the default hazard rate is a nonnegative function of S t and t, (S; t). (S; t) :R + R +! R + We also assume that (S; t) is a decreasing function of S because it seems natural that the probability of default becomes higher when the stock price becomes lower, and vice versa. More specically, noting that the stock price itself is subject to default risk and should satisfy the equation (1) with zero recovery rate(that is, the fractional loss rate is 1), we suppose that S t follows a diusion process under the assumption of a deterministic default-free interest rate: ds = f(s; t)+(s; t)gsdt + Sdw t (3) where r(t), d(s; t), r(t) is a function of the time parameter t, is a positive constant, and d(s; t) denotes the dividend rate which could be a function of the stock price S and the time parameter t. The process can be justied by the fact that the pre-default value of the stock with its cumulative dividend discounted by the default-adjusted rate is a martingale under the risk-neutral measure; that is, e, R t 0 R(u)du S t + Z t 0 e, R u 0 R(s)ds d(s; u)du (4) 7
is a martingale under Q where R(t) =r(t)+(t) because L(t) 1. The assumption that the risk-free rate is deterministic is just for simplicity; clearly, we can easily extend the model to the one in which the risk-free rate is a function of the time parameter and a vector of random variables Y (2 R n ), r = r(y;t), and Y could be described by a multi-dimensional Markov process. In that case, S together with Y also follows a multi-dimensional Markov process and any Markovian term structure model could be combined. Next, we characterize the valuational equation (1) through payos of a CB. If the conversion is allowed only at maturity, X and D t in the equation (1) for CB are given by X = max[as T ;F] and D t = X i c i 1 ftti g (5) respectively where a is a positive constant representing the conversion ratio, F denotes the face value and c i is the coupon payment at time T i. We assume for simplicity that the fractional loss of the market value of a CB is a constant, L(t) =L while we can also dene L(t) as a function of CB's value itself V t, the stock price S t and the time parameter t. In this basic case, a CB is regarded as a non-convertible corporate bond plus a call option on the underlying stock. V t = E Q t " X i R e, Ti R R(u)du t c i + e, T # R(u)du t F 8
+ ae Q t e, R T t R(u)du max[s T, k; 0] (6) where k F ; the rst term and the second term represent the price of a non-convertible corporate bond and the price of a call option on the stock respectively. The CB can be evaluated by standard numerical technique such as Monte Carlo simulations, and in particular, if the dividend rate d(s; t) and the hazard rate (S; t) are non-stochastic, the price is obtained explicitly by utilizing the Black-Scholes formula, a similar formula to the one obtained in the non-defaultable case. V t = " X i R e, Ti t R(u)du c i + e, R T t R(u)du F R +a e, T R fd(u),(1,l)(u)gdu t S t (d 1 ), e, T R(u)du t k(d 1, p T, t) # (7) where d 1 = log St k R + T t f(s)+(s)gds + 1 p 2 2 (T, t) ; T, t and (x) denotes the standard normal distribution function evaluated at x. If the conversion is allowed before maturity, the valuation problem is formulated as an optimal stopping problem: The value of a CB at time t provided that default has not occured by time t is expressed by V t = sup E Q t 2S X + t<t i R e, R R(u)du t 1 f<tg as + e, T R(u)du t 1 f =T g max[as T ;F] 3 R e, T i R(u)du t c i 5 (8) where S denotes the set of feasible conversion strategies which are stopping times taking values in [t; T ]. See chapter 2 of Karatzas-Shreve[1998] for 9
rigorous mathematical argument. Practically, by standard argument as in American options, this can be solved with a recursive backward algorithm in a discretized setting: Starting with V T = max[as tn ;F] at maturity t N T, we implement the following alogorithm at each descritized time point t j, j = N, 1;N, 2; ; 0, where t 0 denotes the date of valuation; V tj = maxfe Q t j [e,r(t j )t V tj+1 ]; as tj g; t t j+1, t j : (9) Further, at each date of the coupon payment T i, V Ti is replaced by V Ti + c j. We can also derive the associated partial dierential equation(pde) by noting that in the equation (1) the pre-default value with the cumulative dividend discounted by the default-adjusted rate is a martingale under Q: 1 2 2 S 2 V SS + f(s; t)+(s; t)gsv S + V t,fr(t)+l(s; t)gv + f(t) =0; (10) where V S, V SS, and V t denote the rst or second order partial derivatives with respect to S or t. f(t) represents the coupon payments, f(t) = P i c i(t, T i ) where () denotes the delta function. In the similar way asin non-defaultable securities, the boundary condtions are given by V (S; T ) = max[as T ;F] at maturity, and V (S; t) as t for conversion when the conversion is allowed before maturity. Further, if there are call and/or put conditions, we add the boundary condtions such as V (S; t) max[cp(t);as t ] in the callable period and V (S; t) pp(t) at the redemption date, where cp(t) and pp(t) denote the call and put prices at time t respectively. See Brenann-Schwartz[1977,1980] and McConnell-Schwartz[1986] for the details 10
of boundary conditions. In practice, we can utilize standard numerical schemes such as nite dierence methods for solving the PDE as in Brenann- Schwartz[1977] of Structural Approach. We emphasize that our PDE unlike the one derived in Tsiveriotis and Fernandes[1998], includes the hazard rate (S; t) in the coecient ofv S because we explicitly take into account that the underlying stock price itself is subject to default risk, which is theoretically more consistent. 3 Numerical Examples 3.1 Implementation and Sensitivity Analyses In this subsection, we briey describe how to implement our model in the subsequent numerical analysis. We take the function of (S t ;t)as (S t ;t)=(s t )= + c S b t (11) where 0, b, and c are some constants. To estimate the credit parts of a CB we utilize information implied in a non-convertible corporate bond(sb) issued by the same company. Notice that the price of a non-convertible bond is consistently evaluated in our framework. That is, in the equation (1), X and D t are specied by the face value and coupon payments respectively. The default-adjsuted short term rate R consists of three parts; the defaultfree short rate r and the hazard rate are common while the fractional loss rates L are generally dierent inconvertible bonds and non-convertible bonds. Practically, we take the following approach; assume that the fractional loss rate of a non-convertible corporate bond denoted by L SB is a 11
constant and is the same as that of a CB, that is L = L SB since it is usually dicult to estimate and L, or L and L SB separately. Then, given, c, and L, estimate b in the equation (11) by calibrating the price of a nonconvertible corporate bond of which maturity is the closest to that of the CB. For computation, we numerically implement a discretized scheme of (9) illustrated in the previous section by utilizing a recombining binomial tree as in Nelson-Ramaswamy[1990] and Takahashi-Tokioka[1999]: Note rst that the process Z t log S t is described as dz t =, 1 2 2 dt + (e Zt ;t)dt + dw t ; Z 0 = log S 0 (12) where (e Zt ;t) (e Zt ;t)+(e Zt ). Then, we discretize the process Z t by N time steps, and approximate it by a binomial lattice in the following; Z(j 0 ;i+1), Z(j; i) =, 1 2 2 h + p hy; Z(0; 0) = log S 0 (13) where h T, Z(j; i) denotes the value of Z at time ih and state j, and N states are characterized by a random variable Y such that j 0 = j + 1 when Y = 1 and j 0 = j when Y =,1 where Y = ( 1 with probability q(j; i),1 with probability 1, q(j; i): (14) Here, q(j; i) is dened so that the expectation of p hy is equal to (j; i)h where (j; i) (e Z(j;i) ;ih). Notice also that the second moment, the expectation of ( p hy ) 2 is automatically equal to 2 h. Hence, we dene q(j; i) with the adjustment for the case that the probability is not in [0; 1]. q(j; i) = 8 >< >: q 0 (j; i); if q 0 (j; i) 2 [0; 1] 0; if q 0 (j; i) < 0 1; otherwise; (15) 12
where q 0 (j; i) = 1 2 1+ p h (j; i)! : (16) Using the binomial scheme and actual market data of corporate bonds, we provide sensitivity analyses of our model. In addition, for comparative purpose, we introduce another new model called boundary model which is similar to the model proposed by Longsta-Shwartz[1995] except in that the underlying state variable is not a rm value, but a stock price. Before showing sensitivity analyses, we describe the structure of boundary model. Boundary Model We suppose that the stock price follows the stochastic process, ds = (S; t)sdt + Sdw t (17) where (S; t) r(t), d(s; t), the risk-free rate r(t) is a function of the time parameter t, the volatility is some positive constant, the dividend rate d is a function of S and t, and w is a standard Brownian motion under the risk-neutral measure Q. Next, dene the default time as = infft : S t S 0 tg: (18) given the critical price of default S 0 (<S 0 ). In the model, the way of recognizing default event is whether the stock price reaches the critical 13
price. We next show that the model also gives a consistent pricing of non-convertible bonds and non-convertible corporate bonds. In this setting, the fundamental partial dierential equation(pde) to price securities issued by a rm is expressed as 1 2 2 S 2 V SS + (S; t)sv S + V t, r(t)v + f(t) =0 (19) where f(t) represents the coupon payments, f(t) = P i c i(t, T i ). For pricing a CB, the boundary condtions are given by V (S; T ) = max[as T ;F] at maturity, V (S 0 ;) = (1, L 0 )(F + P T i c i ) at the time of default, and V (S; t) as t for conversion when the conversion is allowed before maturity, where L 0 denotes the fractional loss rate of the principal and coupons. For pricing a non-convertible bond, the boundary condtions are given by V (S; T ) = F at maturity and V (S 0 ;) = (1, L 0 )(F + P T i c i ) at the time of default. We rst estimate S 0 through the calibration of the market price of a corporate bond of which maturity is the closest to that of the CB in target, and then use that S 0 to price the CB. We also note that if the risk-free rate r(t) and the dividend rate d(s; t) are positive constants, the initial price of non-convertible bond, SB(0) is obtained by the following explicit formula: SB(0) = (1, L 0 )ffa(t )+ X i c i A(T i )g + e,rt FB(T )+ X i e,rt i c i B(T i ); (20) where A(u) S0 f 1 2, (+z) S 0 2 g ", log (S 0=S 0 ), zu p u! S0,2z + S 0 2, log (S 0=S 0 )+zu p u!# ; 14
B(u)! log (S 0=S 0 )+(, 2 =2)u S0 1, 2 p 2, u S 0!, log (S 0=S 0 )+(, 2 =2)u p u and z f(, 2 =2) 2 +2r 2 g 1 2. Here, (x) denotes the standard normal distribution function evaluated at x. For numerical computation of CBs and SBs, we utilize a lattice model with ecient technique for barrier options as in Ritchken[1995]. In the following analyses, r(t) is determined through calibration of the current term structure implied in the Japanese LIBOR and swap market. For the volatility parameter in the stock price process (3), we use the historical volatility computed from the last half-year. We also note that all the CBs used in the analyses have no call nor put conditions. Basic parameters' values and names of CBs used in the analyses are summarized below. =0:001, c =0:6 in the equation (11). L = 1 except in the recovery rate sensitivity analysis (gures (5-a)(5- c)). Sega Enterprise No.4 at 8/11/1999 in the credit spread sensitivity analysis (gures 1 4). In the recovery rate sensitivity analysis (gures (5-a)(5-c)), { Sega Enterprise No.4 at 8/11/1999 for out-of-the-money (OTM), { Nissan Motor No.5 at 11/13/2000 for at-the-money (ATM), { Softbank No.1 at 10/1/1999 for in-the-money (ITM). 15
Nissan Motor No.5 at 11/13/2000 in the volatility sensitivity analysis (gure 6). The following gure shows the hazard rate functions calibrated for the dierent credit spreads of a non-convertible corporate bond with xed parameters except b. Figure 1 W can observed that when the credit spread is low, the hazard rate declines rapidly as stock price rises, which is due to the increase in the calibrated parameter b. The next gure shows the CB prices using these implied hazard rate functions. Figure 2 We notice that when the stock prices reach a certain low level, the rapid decline in CB prices is caused by increasingly rise in the hazard rate. For boundary model, the next gure shows the default boundary S 0 against 16
the dierent credit spreads. Figure 3 Clearly, the default boundary increases as the credit spread is widen. Using the default boundaries for the dierent credit spreads, we compute the CB prices. Figure 4 We can observe that when the stock price is low, CB prices reect the dierence of the credit spreads while there is little dierence in CB prices against the dierent credit spreads when the stock price is high. The following three gures show the recovery rate sensitivity, where the recovery rate is dened by 1, L. We also call our model intensity model in 17
the gures. Figure 5-a,5-b,5-c For our model, we can observe the positive sensitivity of the price against the recovery rate in the at-the-money(atm) and in-the-money(itm) cases while the price is not sensitive to the change in the recovery rate in the out-of-the-money(otm) case. Given the constant credit spread, the hazard rate should rise as the recovery rate increases, which picks up the drift of the stock price process under the risk-neutral measure; as a result, the CB price increases and the eect is the strongest in the ITM case. On the other hand, in the boundary model, the CB price is not senstive to the change in the recovery rate for all cases. The nal gure shows the volatility sensitivity. In the gure, the market price is 129.5, and HV denotes the historical volatility computed from the last half-year. As expected, the CB price rises as the volatility increases for both models. However, in the boundary model, we are not able to nd the 18
default boundary for low levels of the volatility. Figure 6 3.2 Comparison of Models in the Japanese CB Market We next implement comparison of our model and other practical models using actual market data in Japan. We briey illustrate three models used in the analysis other than our model and the boundary model explained in the last subsection, and hereafter call our model Model 1 and the boundary model Model 2: (Model 3)Tsiveriotis-Fernandes [1998] Model In Tsiveriotis-Fernandes[1998] the valuation problem of CBs is formulated as a system of the following two coupled partial dierential equations: 2 S 2 2 u SS + r g Su S + u t, r(u, v), (r + r c )v =0 2 S 2 2 v SS + r g Sv S + v t, (r + r c )v =0 (21) where u is the value of the CB, v is the value of the COCB, S is the price of the underlying stock, r g is the growth rate of the stock, r is the risk-free rate, r c is the observable credit spread implied by nonconvertible bonds of the same issuer for similar maturities with the CB. 19
Here, COCB is dened as follows; the holder of a COCB is entitled to all cash ows, and no equity ows, that an optimally behaving holder of the corresponding CB would receive. See their artice for the details. (Model 4)Goldman Sachs [1994] Model This model modies the binomial pricing model for American options so that the discount rate is adjusted to reect a given credit spread with calculation of the probability of the conversion. we illustrate the part of the adjustment. Let P (S; t) and (S; t) the probability of conversion and the discount rate resepectively when the stock price is S at time t. First, we notice that P (S; t) is determined at maturity T as follows: P (S; T )= ( 1 for as > F 0 otherwise: (22) We next show the essential part of the backward scheme: Let i and j indexes denoting the time and the state respectively. Then, P (i; j) is computed by using P (i +1; ) as P (i; j) =qp(i +1;j +1)+(1, q)p (i +1;j) (23) where q is the risk-neutral probability for the stock to rise in the next time step. Once P (i; j) is obtained, (i; j) can be calculated by using P (i; j): (i; j) =P (i; j)r + f1, P (i; j)g(r + r c ) (24) where r is the risk-free rate and r c is the observed credit spread implied by the non-convertible bonds of the same issuer for similar maturities 20
with the CB. Finally, P (i; j) is changed to 1 if the conversion is optimal; P (i; j) = ( 1 for as > F P (i; j) otherwise: (25) Given P (S; T ), if the one-step scheme is recursively used with the algorithm for American options, the price is obtained. (Model 5)Chen-Nelken [1994] Model Chen-Nelken[1994] is a two-factor model in which factors are a stock price and the yield of a corporate bond. The main assumption is that there is no correlation between the yield and the return on the stock. See their article for the detail of the model. We take Hull-White model as the yield process. Finally, using Japanese CBs'data, we compare our model with other four models explained above. We list below the details of CBs used for the analysis. We note that all the CBs used in the analysis have no call nor put conditions. Sega Enterprise, ltd. CB No.4 All Nippon Airways co., ltd. CB No.5 The Nomura Securities co., ltd. CB No.6 Nissan Motor co., ltd. CB No.5 For calibration we choose a corporate bond of which maturity is the closest to the maturity of each CB. Further, as the indicator for comparison, we 21
use absolute error ratio which is dened by absolute error ratio jmodel price, market pricej : market price The result is listed in the following table. Table 1 We can observe that our model is the best to explain the market data in that the sum of absolute error ratio is the smallest; the absolute error is the smallest for Sega Enterprise and Nomura Securities while that is the third to the smallest for All Nippon Airways and Nissan Motor. Hence, we conrm that our model is relatively valid in practice. 4 Conclusion We have developed a new model for convertible bonds, where we explicitly took default risk into account based on Due-Singleton(1999). The proposed model provides a consistent and practical method for relative valuation of securities issued by a rm such as CBs, non-convertible corporate bonds and equities. In addition, we have shown numerical comparison of our model and other practical models. Finally, we remark that the model can be easily extended to the one in which a risk-free interest rate is described by every Markovian term structure model while it is currently assumed to be deterministic for simplicity. 22
References [1] Brennan, M. and E. Schwartz. `Convertible Bonds: Valuation and Optimal Strategies for Call and Conversion.' The Journal of Finance, 32 (1977), 1699-1715. [2] Brennan, M. and E. Schwartz. `Analyzing Convertible Bonds.' Journal of Financial and Quantitative Analysis, 15 (1980), 907-929. [3] Cheung, W. and I. Nelken. `Costing the Converts.' Risk, July (1994). [4] Due, D. and K. Singleton. `Modeling Term Structure of Defaultable Bonds.' Review of Financial Studies, 12 (1999), 687-720. [5] Goldman Sachs. `Valuing Convertible Bonds as Derivatives.' Research Notes, Quantitative Strategies, Goldman Sachs, November (1994). [6] Ingersoll, J.E.Jr. `A Contingent-Claims Valuation of Convertible Securitis.' Journal of Financial Economics, 4 (1977), 289-382. [7] Jarrow R.A. and S.A. Turnbull. `Pricing Derivatives on Financial Securities Subject to Credit Risk.' The Journal of Finance, March (1995). [8] Karatzas, I. and S. Shreve. Methods of Mathematical Finance. Springer, (1998). [9] Longsta F.A. and E.S. Schwartz. `A Simple Approach to Valuing Risky Fixed and Floating Rate Debt.' The Journal of Finance, 50 (1995), 789-819. 23
[10] Madan D. and H. Unal. `Pricing of Risks of Default.' Working Paper, College of Business, University of Malyland, (1993). [11] McConneLL, J.J. and E.S. Schwartz. `LYON Taming.' The Journal of Finance, 41 (1986), 561-577. [12] Merton, R. `On the Pricing of Corporate Debt:The Risk Structure of Interest Rates.' The Journal of Finance, 29 (1974), 449-470. [13] Nelson, D.B. and K. Ramaswamy. `Simple Binomial Processes as Diusion Approximations in the Financial Models.' The Review of Financial Studies, 3 (1990), 393-430. [14] Nyborg K.G. `The Use and Pricing of Convertible Bonds.' Applied Mathematical Finance, 3 (1996), 167-190. [15] Ritchken, P. `On Pricing Barrier Options.' The Journal of Derivatives, Winter (1995). [16] Takahashi A. and T. Tokioka. `A Three-factor Lattice model for Cross-currency Products', Gendai Finance, March, 5 (1999), 3-16, (in Japanese). [17] Tsiveriotis K. and C. Fernandes. `Valuing Convertible Bonds with Credit Risk.' The Journal of Fixed Income, September (1998), 95-102. 24
0.25 0.2 Implied intensity function (credit spread sensitivity) 0.15 Intensity 0.1 100 bps 200 bps 400 bps (credit spread) 0.05 50 bps 0 0 500 1000 1500 2000 2500 3000 3500 Stock price [Figure 1]
cb price (credit spread sensitivity) 110 cb price 105 100 95 90 85 50 bps 100 bps 200 bps 400 bps 80 10 100 500 1000 2000 3000 5000 stock price [Figure 2]
stock price 500 400 300 200 100 0 Implied default boundary (credit spread sensitivity) 10 50 100 200 400 credit spread default boundary [Figure 3]
cb price 120 110 100 90 80 70 60 50 40 30 20 10 cb price (credit spread sensitivity) 500 1000 2000 3000 5000 stock price 50 bps 100 bps 200 bps 400 bps [Figure 4]
Recovery rate sensitivity (OTM) cb price 100 99.5 99 98.5 98 Intensity Boundary Market price(98.2) 97.5 0 0.1 0.2 0.3 0.4 recovery rate [Figure 5-a] 5
Recovery rate sensitivity (ATM) cb price 131 130 129 128 127 126 125 124 123 0 0.1 0.2 0.3 0.4 recovery rate Intensity Boundary Market price(129.5) [Figure 5-b] 5
Recovery rate sensitivity (ITM) cb price 512 510 508 506 504 502 500 498 496 494 492 0 0.1 0.2 0.3 recovery rate Intensity Boundary Market price(499.8) [Figure 5-c] 5
Volatility sensitivity(hv=0.4969) cb price 160 140 120 100 80 60 40 20 0 0.2 0.3 0.4 0.497 0.6 0.7 0.8 0.9 volatility Intensity Boundary Market price [Figure 6]
(SEGA) (ANA) (NISSAN) (NOMURA) [Table 1: Comparison of Models] Model 1: Intensity model Model 2: Boundary model Model 3: Goldman Sachs (1994) Model 4: Tsiveriotis and Fernandes (1998) Model 5: Cheung and Nelken (1994) date stock HV bond libor bond Market Model Number price price cb bond yield price 1 2 3 4 5 1999/7/5 1761 0.395 99.26 0.29 0.55 2.329 97.9 98.4 99.3 97.7 94 98.4 1999/7/8 1782 0.395 99.07 0.292 0.582 2.415 97.9 98.3 99.2 97.4 94.2 98.1 1999/7/13 1710 0.401 99.15 0.278 0.535 2.382 97.7 98.3 99.3 97.7 94 98.2 1999/7/14 1685 0.402 99.1 0.265 0.532 2.405 98 98.3 99.6 97.7 93.9 98.2 1999/7/26 1676 0.394 99.22 0.247 0.511 2.355 98.3 98.4 99.4 97.8 94.6 98.4 1999/7/30 1610 0.405 98.89 0.274 0.571 2.51 98.4 98.2 99.6 97.7 94 98.4 1999/8/2 1875 0.462 98.85 0.276 0.591 2.53 98.5 98.3 99.7 97.7 92.8 98.4 1999/8/4 2085 0.518 98.76 0.294 0.624 2.573 98.6 98.4 99.8 97.7 91.8 98.5 1999/8/9 2065 0.525 98.82 0.3 0.612 2.548 98.5 98.4 99.5 97.8 91.6 98.5 1999/8/10 1979 0.526 98.71 0.29 0.63 2.601 98.4 98.4 99.5 97.7 91.2 98.5 1999/8/11 1940 0.526 98.8 0.3 0.62 2.559 98.2 98.4 99.5 97.7 91.1 98.5 1999/8/18 1935 0.53 98.82 0.298 0.622 2.554 98.4 98.4 99.8 97.8 91.3 98.5 1999/8/31 2000 0.529 99.12 0.282 0.597 2.419 98.5 98.6 99.7 98 92 98.7 1999/9/13 2290 0.546 99.25 0.25 0.53 2.363 96.5 98.7 99.9 98.3 93 98.9 1999/10/13 1998 0.574 99.56 0.22 0.438 2.221 96.4 98.5 99.3 98.2 91.6 98.8 1999/11/12 1900 0.564 99.71 0.22 0.41 2.151 96.4 99 99.9 98.4 92.4 99.1 1999/12/27 3230 0.754 99.61 0.22 0.4 2.217 97.8 100.7 101.5 100.2 95.9 100.9 sum of error ratio 0.131 0.308 0.153 0.867 0.134 1999/7/15 373 0.354 98.88 1.495 1.667 2.222 88.3 92.3 88.3 92.3 82.2 95.4 1999/8/30 397 0.354 95.87 1.69 1.902 2.709 86.2 91.3 88.1 90.6 81.8 95.7 1999/9/10 381 0.318 96.96 1.54 1.746 2.534 86.1 90.1 89.6 90.4 81.7 95.6 1999/10/22 341 0.285 96.54 1.431 1.763 2.611 88.1 89.8 86.7 89.3 81.8 93.3 1999/11/11 327 0.267 97.58 1.615 1.785 2.443 87.6 90.6 86.2 89.9 82.2 93.7 1999/12/3 323 0.269 98.03 1.345 1.506 2.371 89.4 90.6 84.7 90.3 82.7 93.6 sum of error ratio 0.218 0.147 0.196 0.379 0.477 1999/7/1 1425 0.201 94.56 0.986 1.876 2.874 100 99.9 97.7 95.1 94.4 99.7 1999/7/19 1774 0.208 94.83 0.941 1.858 2.842 100.1 100.4 96.9 95.6 95.7 101.9 1999/8/11 1490 0.215 94.91 1.035 2.082 2.835 98.9 99.8 97.9 95.6 94.8 100.6 1999/8/23 1803 0.223 94.58 1.023 2.125 2.881 100.6 100.8 98.5 95.8 95.8 102.2 1999/9/13 1676 0.218 95.94 0.914 1.891 2.704 99.4 99.7 97.3 96.4 96 102 1999/10/15 1740 0.213 96.36 0.779 1.907 2.653 99.8 99.5 97.7 95.3 95.5 101.5 1999/11/11 1730 0.211 97.34 0.728 1.976 2.526 99.9 99.6 96.5 95.9 96.1 101.9 2000/1/24 2400 0.235 99.49 0.745 1.702 2.247 102.1 100.3 100.7 98.8 99.6 105.5 2000/2/4 2995 0.243 98.71 0.79 1.802 2.351 108 104.4 102.7 101.6 102.6 110.1 sum of error ratio 0.075 0.225 0.382 0.379 0.17 1999/7/5 582 0.453 99.19 0.879 0.871 2.405 108.8 119.8 115.5 119.6 122.1 125.7 1999/9/30 645 0.387 100.1 0.701 0.691 2.147 116.9 122.4 116 120.6 124.4 126.5 1999/10/4 690 0.393 100.2 0.825 0.818 2.117 127.7 125.6 120 124.4 128.8 130.5 1999/11/4 624 0.397 100.6 0.807 0.8 1.992 116.2 120.4 116.8 119.7 123.4 125.1 1999/11/16 520 0.462 100.6 0.8 0.793 2.003 104.6 115.6 109.7 114.9 116.4 119 1999/12/27 405 0.525 100.4 0.776 0.768 2.042 98.5 110.5 104.2 109.8 107.4 112.8 2000/11/2 760 0.507 101.3 0.715 0.711 1.618 128 134.2 130.6 133.7 136.2 136.5 2000/11/13 720 0.497 101.3 0.709 0.705 1.598 129.5 129.6 126.2 129.1 132.1 132.1 sum of error ratio total sum of error ratio 0.477 0.287 0.448 0.544 0.705 0.901 0.968 1.179 2.17 1.486